Re: [Idr] Progressing draft-bhatia-bgp-multiple-next-hops-01.txt

["John G. Scudder" <jgs at cisco.com>]

A few other points regarding the draft --

First, the draft isn't terribly clear on the usage of "legacy" 
next_hop on a connection over which the multiple-hop capability has 
been exchanged.  To take one example, if I announce:

    R1: (P, next_hop N1, other attributes)
    R2: (P, multiple_next_hop N2, other attributes)

can I subsequently withdraw R1 by means of an update which lists N1 
in the multiple_next_hop attribute?  I'd assume so, but the mixing of 
next_hop and multiple_next_hop leaves room for confusion.  (Other 
comments to the list have exhibited the same confusion so I think 
it's not just me.)

I'd suggest dropping use of next_hop at all on multiple-hop 
connections, for clarity.  Instead, mandate that all next hop 
information is carried in multiple_next_hop.  As far as I can tell, 
the only well-defined use of next_hop in the document right now is if 
you want to implicitly withdraw N routes and replace them with M 
routes (but where all M share one set of path attributes).  That 
seems like an unnecessary corner case optimization, especially since 
you've already disclaimed interest in other implicit withdraw cases 
(where the next hop changes).

If you don't want to make this change then I think you at least need 
to clarify the implicit withdraw semantics in the presence of 
next_hop, as Paul did --

At 5:26 PM +0100 8/23/06, Paul Jakma wrote:
>    The UPDATE replaces all existing paths for the NLRI with the
>    paths of identities (NLRI,next-hop) as given by the
>    'normal' next-hop attribute and those nexthops in the
>    MULTIPLE_NEXT_HOP, if that attribute is present.

This is NOT clear in the doc.

Second, the draft specifies a set of flags in the multiple-hop 
capability.  The only defined flag is the "receive multiple" flag. 
Under what circumstances would a router ever advertise the capability 
without setting the RM flag?

Third, you misspelled "Traina" in your references section.

--John

_______________________________________________
Idr mailing list
Idr at ietf.org
https://www1.ietf.org/mailman/listinfo/idr